When you buy through links in our articles, Future and its syndication partners may earn a commission.
If we were to take a scenic drive through each era of astronomy, we would begin somewhere among the cavemen wondering why there are stationary fireflies in the sky. We would travel through libraries of scrolls about how those fireflies are really siblings of our giant yellow sun, and then through rooms of books about how our entire world somehow revolves around that sun.
Ultimately, scientists discovered that gravity depends on the structure of space-time, imaged iridescent galaxies other than the Milky Way, and calculated the strict limits of supermassive black holes.
But as we approach our exit to the present, I think we’re going to see something really interesting. We’re going to see the growing bond between astronomer and machine, which is going to allow us to open cosmic doors more quickly. Aritra Ghosh, a postdoctoral researcher at the University of Washington, is one of those astronomers.
For example, Ghosh recently managed to confirm that galaxies in denser regions of the universe can be up to 25% larger than galaxies of similar mass and shape in less dense regions. “Size” in this case refers to the radius of a galaxy that contains 50% of its total light output. This is a nice result in itself, but it’s especially important to emphasize how it was achieved: by using machine learning to study more individual galaxies than the human body could analyze in a single lifetime. To be precise, there were 2,894,716 galaxies in the dataset.
“Over the past decade, many astronomers, like myself, have done painstaking studies to build confidence in machine learning, showing that it can replicate traditional techniques,” Ghosh told Space.com. “We can finally start using these techniques to get new scientific results.”
This massive sample of galaxies actually came from an even larger set that Ghosh obtained using machine learning. That original set, obtained with a research tool called GaMPEN, included data around 7,805,186 galaxies — the smaller subset for this new study was selected based on where the galaxies lie in the sky. In a single millisecond, GaMPEN can infer the structure of a galaxy based on a parameter the user chooses; Ghosh and his fellow researchers used a parameter that revealed what fraction of the light comes from a galaxy’s outer disk compared to its central bulge.
“I wanted to demonstrate to the broader community how machine learning and large image datasets can be combined to make progress on long-standing questions in astrophysics,” said Ghosh.
Ghosh then took those nearly 8 million subjects from regions where he knew the density of the universe from previous calculations. In the work, “dense” environments include many things, including regions where you would find superclusters of galaxies. These are giant conglomerates of many galaxy clusters (a single galaxy cluster can contain up to 1,000 individual galaxies!) that are typically found in the strands of the cosmic web that permeates our entire universe. You can think of them as the hot spots at the center of the universe.
“Our collaborators in Japan, led by Rhythm Shimakawa, measured the ambient densities,” Ghosh said. “They used a non-ML computer algorithm to place circles with a radius of 30 million light-years in different parts of the sky and count the number of galaxies inside each circle — circles in denser regions have a higher than average count.”
Once the subgroup was identified, Ghosh and his team began looking at correlations between the size of the galaxy and its environment.
Because a galaxy’s mass is strongly related to its size and environment — for example, larger galaxies are expected to grow larger and live in denser environments — the team compared the sizes of galaxies of the same mass in different environments. “Since massive galaxies are rare,” Ghosh explained, “we collaborated with theoretical astrophysicists to develop a new metric for the correlation analysis.”
Moreover, not only is this the largest catalog ever used to study the size and environment of a galaxy – and, Ghosh speculates, probably one of the top five astrophysical studies – but it also has an error-correction mechanism that Ghosh says has been largely absent from previous similar studies, thanks in part to the machine learning component.
Speaking of those previous studies, the finding that larger galaxies are more likely to be found in supercluster cities than in rural cosmic villages was a bit of a surprise — despite sounding relatively intuitive. As Ghosh explains, many scientists who have studied the ins and outs of galaxies in clusters believed that strong dynamical forces within those clusters would gradually strip matter from a galaxy, causing it to shrink.
But the team saw larger galaxies in dense, supercluster environments. Weird.
“We tested our correlation algorithm on smaller subsets first,” Ghosh said. “The ‘Aha!’ moment was when we ran the analysis for the first time on the full sample of 3 million galaxies, and noticed the strong positive correlation.”
Why might this be? Well, there are a few possibilities. One has to do with the kind of “matter” that is suggested to be shed from galaxies in dense regions of the universe — normal matter, made up of standard protons, neutrons, and electrons. This begs the question: what about dark matter? Perhaps this invisible substance plays a role in keeping galaxies larger. It wouldn’t be a far-fetched idea, since scientists have shown that most large galaxies are surrounded by a halo of dark matter, including our own Milky Way.
“Our work shows that when you average over many clusters, dark matter becomes the primary driving force, reversing the trend we see in individual clusters,” Ghosh said.
However, it is also possible that galaxies in denser environments are larger when they first form. Furthermore, there is a chance that denser environments increase the likelihood and ease of galactic mergers.
“An interesting follow-up work would be to check how this result changes if you change the radius of the circle in which you measure the density,” Ghosh said. “What if you use a radius of 1 million light years instead of 30? This will tell us how physics at different scales of the universe affects galaxies differently.”
In the meantime, the team has its eyes on the Rubin Observatory, which is expected to see first light from the cosmos in early 2025, and the massive data sets it should produce.
Related Stories:
— James Webb Space Telescope Discovers ‘Cosmic Gems’ in Extremely Early Universe (Video)
— Huge galaxy cluster ‘El Gordo’ suggests dark matter is colliding with itself
— World’s largest visible-light telescope sees a cluster of galaxies that warps space-time
“My current fellowship focuses on the Rubin Observatory,” Ghosh said, “which will observe 20 billion galaxies over its lifetime.”
And even if Rubin somehow manages to find a few extra puzzle pieces under the couch instead of putting a few on the table, there is still a tangible success to Ghosh’s study. It is evidence that machines can be trusted to answer questions about the universe we’ve brought them into.
The study was published on August 14 in The Astrophysical Journal.